Keywords: Computational Intelligence, Evolutionary Computation, Metaheuristic Algorithms
Abstract: Structure

1. Introduction to Combinatorial Problems [20 min] 2. Systematic Search Techniques [30 min] 3. Metaheuristics [20 min] 4. Change, Uncertainty, and Preferences [20 min] 5. Machine Learning for Modeling and Solving [20 min] 6. Conclusion [10 min] Outline

Combinatorial applications, such as timetabling, resource allocation, scheduling, and planning, consist of finding a good / best consistent scenario satisfying a set of constraints while optimizing some objectives, including user preferences. In addition, constraints and objectives might not be explicitly defined and often come with uncer-tainty due to lack of knowledge, missing information, or variability caused by external events. Finally, in some applications such as, these constraints and objectives can be temporal, spatial, or both. In the latter case, we are dealing with entities that occupy a given position in time and space.

The aim of the tutorial is to provide attendees with the necessary tools to address the above challenges when tackling a combinatorial application. We will present a general approach that covers both systematic search and metaheuristics. We will show how systematic search is enhanced, in practice, via constraint propagation and variable ordering heuristics. Metaheuristics will be described through a set of trajectory-based and population-based nature-inspired techniques. We will consider cases where constraint problems occur in dynamic environments and situations where some of the relevant information is incomplete/uncertain. Finally, to deal with requirements and desires that are not explicitly defined, we will explore constraint and preference learning algorithms.

Keywords: Autonomous Vehicle, Cyber-physical systems, Distributed Intelligent Systems
Abstract: Structure

1. Introduction 2. Deep Dive into UAV Applications 3. Crafting the Future: Design Guidelines and Platforms 4. Concluding Remarks and Interactive Discussions

Outline

Unmanned Aerial Vehicles (UAVs) have emerged as a transformative force in technology, capturing the attention of industries, government agencies, and academia. Supported by the National Science Foundation (NSF), our research initially spanned from 2017 to 2022 under a major NSF project and has now entered its second phase with a new award starting in 2023. Despite advancements in UAV control, communication, networking, and computing, fully unlocking the potential of airborne computing remains a significant challenge. This tutorial addresses this gap, laying the foundation for a new era of UAV-centric airborne computing.

This tutorial will: (1) explore current and emerging UAV applications, analyzing their complexities; (2) present real-world case studies demonstrating how airborne computing transforms UAV functionalities; (3) provide essential design strategies for next-generation UAV systems enhanced by airborne computing; (4) showcase our cutting-edge UAV-based airborne computing platform and latest prototype; and (5) explore pioneering UAV functions, encompassing reinforcement-learning guided antenna positioning, coding-driven distributed computing and federated learning, software-defined radio-powered cellular base stations, and deep-learning-enhanced object detection.

Aligned with IEEE SMC 2025’s theme, this tutorial highlights the pivotal role of airborne computing in advancing UAV design and enabling innovative applications. Attendees will leave with a deeper understanding of the challenges, innovations, and future opportunities in UAV-based airborne computing.

Keywords: Evolutionary Computation, Computational Intelligence, Metaheuristic Algorithms
Abstract: Structure

This tutorial follows a structured and interactive format designed to provide participants with a comprehensive understanding of Evolutionary Multi-Objective Optimization (EMO). The session will be divided into the following key sections:

1. Introduction to multi-objective optimization 2. Overview of evolutionary algorithms 3. Basic concepts of evolutionary multi-objective optimization (EMO) 4. General categories of EMO algorithms 5. Designing EMO algorithms 6. Applications of EMO algorithms 7. Recent developments and hot topics 8. Conclusions Outline

Multi-objective optimization problems are commonly found in many real-world applications. These problems involve the simultaneous optimization of multiple conflicting objective functions. Consequently, they do not yield a single optimal solution but rather a set of trade-off solutions. The trade-off solutions are often defined using the Pareto dominance relation and are known as Pareto optimal solutions. When mapped to the objective space, they form the Pareto front. Evolutionary Multi-Objective Optimization (EMO) algorithms have been a popular approach for solving multi-objective optimization problems. Thanks to their population-based search nature, EMO algorithms can obtain a set of non-dominated solutions in a single run, which is then used to approximate the Pareto front. This tutorial will give a comprehensive introduction on the fundamental concepts of evolutionary multi-objective optimization including commonly used strategies in designing EMO algorithms. Additionally, in this tutorial, we will cover some recent hot topics and advancements in the field, ensuring audiences are updated on the latest developments in evolutionary multi-objective optimization.

Keywords: Networking and Decision-Making, Visual Analytics/Communication, Information Systems for Design
Abstract: Large-scale low Earth orbit (LEO) satellite networks are characterized by massive node populations, rapid topology changes, and resource-constrained individual satellites. Existing routing technologies in such networks face challenges such as slow convergence, excessive end-to-end hop counts, and prolonged latency. To address these issues in applying Q-learning to large-scale LEO constellation routing, this study proposes a Maximum Trusted Distance-Based Greedy Q-Learning Routing (MTD-GQR) strategy. The approach first optimizes Q-tables through a greedy mechanism to accelerate convergence. It then extends beyond traditional topological constraints, where satellites link only with four adjacent neighbors, by leveraging maximum trusted distance to define inter-satellite connectivity. Simulations comparing the proposed strategy with the Dijkstra algorithm and recent Q-learning-based approaches for large-scale constellations demonstrate significant improvements: a 26.7% reduction in convergence time, a 15.8% decrease in average end-to-end hop count, and a 3.1% reduction in average latency. These results highlight the effectiveness of MTD-GQR in enhancing routing performance for large-scale LEO satellite networks.

Keywords: Design Methods, Human Factors, Human-centered Learning
Abstract: When language barriers exist, sketching as a form of non-verbal visual expression can facilitate cross-cultural communication. However, many people lack sketching skills, especially the ability to quickly convey visual ideas. To address this, we conducted experiments collecting eye movement data from individuals with varying sketching experience as they imagined and drew object shapes. Based on this data, we built a classification model using a Bidirectional Long Short-Term Memory network (Bi-LSTM) and a Key Attention Network (KAN). The model achieved a validation loss of 0.0305, an accuracy of 0.90, and an F1 score of 0.8845, demonstrating strong classification performance and generalization on imbalanced data. This study integrates AI with artistic expression, offering intelligent feedback tools for sketch learners and helping non-professionals improve visual communication. Additionally, it provides new insights for art education, cross-cultural interaction, and visual cognition research.

Keywords: Human-Machine Interaction, Wearable Computing, Human Performance Modeling
Abstract: With increasing public attention to health and wellness, treadmill exercise in gyms has become increasingly popular. However, the risk of falling due to loss of balance during treadmill use remains a significant safety concern. To address this issue, we propose a gait stability prediction framework that leverages a deep learning neural network BiLSTM-KAN trained on multimodal data. Specifically, we use MediaPipe to extract 3D skeletal key points and smart insoles to capture essential plantar features and then perform data preprocessing to synchronize the postural features with the plantar data. The trained model is capable of predicting 15 biomechanical indicators related to foot stability based on pose data obtained from a camera, thereby enabling indirect assessment of fall risk without requiring the user to wear any devices during use. Experimental results demonstrate that the proposed system achieves high accuracy in predicting key stability-related parameters such as Gait-line CPE and Impulse. This study validates the feasibility of a non-contact, continuous monitoring approach based on multimodal fusion for gait stability assessment, providing a practical solution for safer exercise and intelligent health monitoring.

Keywords: Communications, Robotic Systems
Abstract: Telepresence technology will enable a revolutionary collaborative model where humans and robots work together, leveraging their unique strengths to extend humanity’s reach into space during this New Era of Space Exploration. The telepresence operations framework allows humans to apply their skills and expertise to complex problems remotely while utilizing the resilience of robots. Telepresence can be applied in various space operations, creating the impression that crew members and ground personnel are present in a different location.

In this full-day Workshop, we will host experts in different aspects of telepresence for space exploration, from communications, to AI, to robotics who will give invited talks and provide their advice and guidance in an expert panel, followed by technical talks from other experts who submitted papers to the Workshop and were peer-reviewed and accepted for publication. We will conclude with a round-table discussion on the roadmap to enable this New Era of Space Exploration within the next 20-30 years.

More information and a detailed program can be found at https://sites.google.com/view/smc-2025-telepresence-workshop/program

Keywords: Shared Control, Haptic Systems, Human-Machine Interface
Abstract: The IEEE SMC Models, Patterns, and Assessment Methodologies: An Interactive Workshop on Shared and Cooperative Control Systems will be held from October 5-8, 2025, as part of the SMC 2025 flagship annual conference of the IEEE Systems, Man, and Cybernetics Society. The goals of the Workshop are to reinvigorate the community, building upon the past successful workshops at SMC in 2012, 2013, and 2015, and to provide a forum to exchange insights and explore the latest advancements in human-machine systems. This workshop offers valuable insights, bringing together experts and practitioners to explore explainable methods, assessment models, and application-specific insights.

More information and a detailed program can be found at https://sites.google.com/view/ieee-smc-sc-workshop/workshop-program

Keywords: Shared Control, Human-Machine Interface, Haptic Systems
Abstract: In haptic shared control (HSC), a human operator and an autonomous controller share a common, physical control interface of a robot. While stronger haptic guidance is beneficial for better safety and task performance, it is problematic when the autonomous controller is unreliable and frequent disagreements between humans and machines occur. While allowing the operator to adjust the haptic guidance strength can alleviate the issue, determining the appropriate guidance strength remains a challenging task. This study proposed a method of utilizing a grip mechanism for the operator to adjust the haptic strength and for the autonomous controller to recommend the appropriate strength based on its reliability by applying proportional control to the grip angle. An experiment using a simulation showed that the proposed method is effective in aiding the adjustment of haptic guidance strength and marginally reducing operator workload.

Keywords: Human-Machine Interaction, Human-Machine Cooperation and Systems, Human Factors
Abstract: Automation is defined as the replacement of tasks previously handled by humans with machines. One value of automated driving is that it enables humans to engage in non-driving tasks, thereby allowing for more efficient use of time. However, humans may suffer from boredom created by monotonous environment or drowsiness due to human characteristics in conditional driving automation, where the role of resuming DDTs is assigned to humans. Humans may have an intention of arriving at their destination faster. Humans may underestimate risks associated with drowsiness and may not form an intention to take a break. Humans may have their attention focused on non-driving tasks and fail to find an opportunity for taking a break. The automated driving system should not continue to vehicle control in a situation where humans, who are expected to act as a backup, feel drowsy. One technology that is currently lacking is a mechanism for humans and machines to communicate their intentions and form joint decisions for strategic, tactical, and operational levels. The mechanism is called mediation, meaning means of interaction strategies aimed to reach a joint intent and an adequate action. Against this background, the purpose of the study is to design human-machine mediation for joint intention formation in coping with drowsiness under driving automation. We constructed a state description of human-machine interaction based on relationship between environment, human intentions and behaviors, and machine intentions and behaviors. As shown in Fig. 1, the flow of interactions that make a joint intent and an adequate action for responding to driver drowsiness was described using Swimlane Diagrams. The Karolinska Sleepiness Scale was used to determine sleepiness levels, and opportunities for human-machine interaction were constructed according to sleepiness levels and levels of automation. In an era of increasing autonomy, technologies that support consensus-building between humans and machines are important, and the contribution was to present a method for describing the state of human-machine interactions, using the use case of coping with drowsiness. We plan to evaluate the usability of the designed mediation in future experiments.

Keywords: User Interface Design, Telepresence, Shared Control

Keywords: AI and Applications, Application of Artificial Intelligence
Abstract: The decision to form a working group was made at the AI Workshop held by the IEEE SMC in February in the presence of BoG members and ExCom members. The workshop organizer was asked to form the working group, and based on the opportunities considered, one of the first active discussions could be at the SMC conference, where most of the members of the working group will participate, and more articles are expected at the conference that deal with the raised problem. The first discussion of the formative IEEE SMC working group on Ai teaching and Ai in teaching will take place, which will, among other things, formulate the main challenges and make suggestions for solutions and suggestions for solutions based on good examples. The workshop can be a starting point for organizing the next actions, formulating standards and involving AI professionals in outlining AI education and its application in education.

Keywords: Intelligent Internet Systems, Cybernetics for Informatics, Cloud, IoT, and Robotics Integration
Abstract: Structure

The agenda is divided into three main parts, each providing opportunities for Q&A and discussions to engage participants and reinforce key concepts. The schedule includes the start and end times, duration, and a brief description of each session.

Part 1, “Welcome and Overview”: This introductory session provides an overview of the tutorial’s content and a brief introduction to the speakers.

Part 2, “Integration of UAVs into wireless networks”: This section begins with an overview of UAV applications in wireless networks (e.g., aerial communication relays for power line inspection and monitoring). It then examines UAVs from a robotics perspective, covering their dynamic models, operational principles, and control algorithms. A general mathematical framework for communication-aware robotics problems is introduced, followed by several use cases that demonstrate the effectiveness of the discussed techniques. The section concludes with a discussion on the limitations of traditional UAV platforms in communication networks, focusing on the inherent coupling between their orientation and motion.

Part 3, “f-MRAVs in Robotics and Communications”: This section introduces fully actuated Multi-Rotor Aerial Vehicles (f-MRAVs) and their key applications. It details their mathematical models and robotics characteristics, highlighting similarities and differences compared to standard UAV platforms. The discussion then focuses on how f-MRAVs can redefine UAV wireless networks through three key applications: (i) enhancing physical layer security, particularly in countering jamming and eavesdropping attacks; (ii) integrating f-MRAVs with antenna arrays and reconfigurable intelligent surfaces (RIS); (iii) supporting integrated sensing and localization. The tutorial concludes with closing remarks, outlining future research directions and facilitating an open discussion with attendees. This interactive session will offer participants the opportunity to engage with the presenters, clarify concepts, and explore new ideas. Outline

This two-hour tutorial delves into the integration of MRAVs into wireless networks, an emerging research area driven by the demand for more flexible and efficient communication systems. Traditional underactuated MRAVs (u-MRAVs), such as quadrotors, exhibit limited control capabilities, resulting in suboptimal performance in dynamic environments and increased vulnerability to threats such as eavesdropping and jamming. To overcome these challenges, f-MRAVs have been developed, offering precise control over all movement axes and significantly enhancing maneuverability within wireless networks. During the session, we will explore how f-MRAVs can redefine wireless communication systems by comparing their performance with conventional u-MRAVs. Topics include the inherent limitations of u-MRAVs, the advanced capabilities of f-MRAVs, and their consequent impact on network performance. We will also discuss the innovative mechanical designs, sophisticated control algorithms, and modeling techniques that underpin f-MRAV technology. The tutorial further outlines future research directions for integrating f-MRAVs into next-generation 6G networks, paving the way for more advanced and secure communication systems. By addressing these state-of-the-art developments, the session underscores the convergence of robotics, control theory, and cybernetic systems.

Keywords: Intelligent Power Grid, Infrastructure Systems and Services, Trust in Autonomous Systems
Abstract: Structure

1. Introduction: AI and the Energy Transition 2. Emerging Trends in AI-Driven Energy Systems 3. Energy Justice Framework 4. Advancing Energy Justice with AI 5. Case Studies (2) 6. Bridging Energy Laws and AI Regulations for a Just Energy Transition 7. Final Q&A & Discussion Outline

The transition to sustainable energy systems presents both technical and socio-economic challenges, particularly in ensuring equitable access to clean energy resources. Artificial Intelligence (AI) is increasingly recognized as a transformative tool in optimizing power systems, enhancing grid resilience, and enabling decentralized energy markets. However, the deployment of AI-driven energy solutions must consider fairness, transparency, and inclusivity to avoid reinforcing existing energy inequities.

This tutorial explores the intersection of AI, energy transition, and justice, with a focus on power systems and cybernetics. It begins with an overview of AI applications in smart grid optimization, energy efficiency, and distributed energy trading. Next, it introduces the Energy Justice Framework, highlighting distributive, procedural, and recognition justice in AI-powered energy systems. Case studies, including AI-driven community energy storage and reinforcement learning for fair load shedding, will illustrate real-world applications. Finally, the tutorial examines regulatory considerations for bridging energy laws with AI ethics to ensure a just energy transition.

Keywords: Visual Analytics/Communication, Human-Centered Transportation, Human Factors
Abstract: Rearview mirrors provide critical driving information, especially in the process of departing from a parking space and merging onto the main road (DPMM condition) . However, existing research has not sufficiently explored how the interaction between driving experience and gender influences rearview mirror observation behavior during this process. This study investigates real-world driver observation behavior toward rearview mirrors. Eye movement data were collected from 10 participants using Tobii Pro Glasses 3 while they performed vehicle departure and merging tasks on a closed road under the DPMM condition. Fixation Count (FC) was used as the primary evaluation metric. Results from a two-way ANOVA revealed a significant interaction effect between driving experience and gender for the right-side mirror. Novice female drivers exhibited a higher fixation frequency, whereas experienced male drivers showed increased fixation counts with more driving experience. These findings suggest that gender and driving experience interact to influence mirror-checking strategies under the DPMM condition. By addressing individual differences in observation behavior, this study contributes to personalized driver training and improved road safety.

Keywords: Cognitive Computing, Human-Machine Interaction, Intelligence Interaction
Abstract: In modern lives, the rapid proliferation of Internet of Things (IoT) devices has made them indispensable tools for data collection and analysis across various domains. However, growing concerns over data ownership and privacy have hindered effective data sharing among IoT devices, leading to the persistent challenge of data silos. Federated Learning (FL) has emerged as a promising solution to this problem by enabling collaborative model training without direct data exchange. Despite its potential, FL faces two critical limitations: severe catastrophic forgetting for historical knowledge and inefficient average aggregation. To address these challenges, this paper proposes FedGA, an innovative FL framework that leverages cosine similarity-based weighted aggregation to enhance model convergence speed. Furthermore, FedGA incorporates a mechanism to memorize historical models, thereby significantly alleviating catastrophic forgetting. Extensive experiments on three public datasets validate the effectiveness of FedGA, demonstrating its superior performance in both accuracy and training efficiency compared to state-of-the-art methods. The results highlight FedGA’s capability to overcome the key shortcomings of existing FL approaches, making it a robust solution for practical IoT applications.

Keywords: Machine Learning, AI and Applications
Abstract: The proliferation of Internet of Things (IoT) devices has introduced significant security vulnerabilities, particularly in detecting zero-day attacks within highly dynamic and heterogeneous environments. Traditional machine learning models often fall short due to their static nature and computational demands. In this paper, we propose an adaptive ensemble learning framework that integrates both supervised and reinforcement learning methods, Random Forest (RF), K-Nearest Neighbors (KNN), Support Vector Machine (SVM), and Q-learning (RL). Additionally, we propose using four ensemble learning techniques, Bagging, Boosting, Stacking, and Voting, for intrusion detection in IoT networks. Ensemble techniques such as Bagging, Boosting, Voting, and Stacking. The main contribution of this paper is in class-wise model selection, optimizing intrusion detection performance per attack type. Finally, we propose a decision-rule mechanism that selects the best-performing model for each attack class to improve detection accuracy. The proposed framework is evaluated through extensive experiments. The results show that our approach significantly enhances classification performance, especially for complex and rare attack types, while ensuring scalability and low computational cost.

Keywords: Cybernetics for Informatics, AI and Applications, Machine Learning
Abstract: Software-defined Networks (SDNs) offer flexibility and programmability but introduce new attack surfaces, especially at the edge. Man-in-the-Middle (MitM) attacks on the control channel between the controller and edge devices pose a significant threat. This paper proposes a lightweight approach for detecting MitM attacks in SDN-based edge architectures. We evaluate several machine learning classifiers based on their effectiveness and efficiency for resource-constrained edge environments. Our approach analyzes real-time network traffic based on features sensitive to MitM activities. Through a simulated case study, we compare the performance of a rule-based baseline against various classifiers, including Gaussian Naive Bayes, Logistic Regression, Decision Tree, Random Forest, LightGBM, and XGBoost. The results demonstrate that the Decision Tree classifier achieves excellent performance, with 99% accuracy, precision, and recall, and critically, exhibits one of the lowest detection latencies (0.7 ms). This balance of high detection capability and low computational overhead positions Decision Trees as a highly suitable lightweight solution for real-time MitM attack detection in SDN edge architectures, outperforming simpler methods like Gaussian Naive Bayes and the rule-based baseline, and offering a more efficient alternative compared to complex ensemble models when latency is a primary concern. We also analyze the relative CPU and memory consumption, further supporting the practicality of Decision Trees for edge deployment.

Keywords: Machine Learning, Information Assurance and Intelligence, AI and Applications
Abstract: Adversarial examples can represent a serious threat to machine learning (ML) algorithms. If used to manipulate the behaviour of ML-based Network Intrusion Detection Systems (NIDS), they can jeopardize network security. In this work, we aim to mitigate such risks by increasing the robustness of NIDS towards adversarial attacks. To that end, we explore two adversarial methods to generating malicious network traffic. The first method is based on Generative Adversarial Networks (GAN) and the second one is the Fast Gradient Sign Method (FGSM). The adversarial examples generated by these methods are then used to evaluate a novel multilayer defense mechanism, specifically designed to mitigate the vulnerability of ML-based NIDS. Our solution consists of one layer of stacking classifiers and a second layer based on an autoencoder. If the incoming network data is classified as benign by the first layer, the second layer is activated to insure that the decision made by the stacking classifier is correct. We also incorporated adversarial training to further improve the robustness of our solution. Experiments on two datasets, namely UNSW-NB15 and NSL-KDD, demonstrate that the proposed approach increases resilience to adversarial attacks.

Keywords: Cybernetics for Informatics, Deep Learning, Neural Networks and their Applications
Abstract: Side-channel attacks and deep learning have both attracted significant attention in recent years. As deep learning continues to advance, its application in side-channel analysis has opened new possibilities for cryptographic key recovery. This paper presents a clear approach for recovering AES keys from side-channel traces using deep learning and introduces an efficient Neural Architecture Search (NAS) method based on random search to enhance the performance of standard Convolutional Neural Networks (CNNs) in side-channel analysis. Specifically, we apply NAS to the VGG16 and ResNet18 architectures, resulting in two optimized models, referred to as NAS-VGG and NAS-ResNet. Our method significantly reduces training time, achieving improved performance after only 10 epochs, compared to the original models, which required 50 or more epochs. Furthermore, the NAS-VGG model not only outperforms the original VGG16 in terms of guessing entropy (GE), but also surpasses a well-established CNN-based approach from the literature, reaching GE=0 with as few as ~450 traces. These results demonstrate the effectiveness of random search-based NAS in discovering compact, high-performing architectures with minimal computational cost.

Keywords: Intelligence Interaction, Human-Computer Interaction, Human-centered Learning
Abstract: Structure

This tutorial will be divided into four sections, each covering a critical aspect of Socratic AI tutoring for K12 math. The session will include a combination of lectures, hands-on activities, and discussions to provide attendees with both theoretical foundations and practical experience.

Outline

As AI-driven educational systems advance, the challenge lies in designing human-machine interactions that are pedagogically effective, adaptive, and socially responsible. This tutorial introduces Socratic AI Tutoring, a structured AI-human conversation framework for K12 math education that engages students in interactive, step-by-step guidance to understand and correct their mistakes—rather than merely receiving answers. By leveraging pre-cached error attribution from draft analysis, AI tutor is already aware of where and why students struggle before initiating dialogue, enabling personalized and targeted interventions. This tutorial is highly relevant to IEEE SMC 2025, aligning with Human-Machine Systems and Cybernetics by showcasing how AI can facilitate structured, adaptive learning through real-time interaction and reasoning. Our system employs a systematic evaluation framework, assessing AI-led conversations across six pedagogical dimensions, and integrates robust control mechanisms to ensure safe and meaningful dialogue. With automated conversation scoring, moderation filters, and adaptive engagement strategies, this framework exemplifies how cybernetic principles can be applied to AI-based education. Attendees will gain insights into AI-driven Socratic tutoring, error attribution, conversational assessment, and moderation techniques, contributing to the future of adaptive, intelligent learning systems. This tutorial is ideal for researchers, educators, and practitioners interested in human-machine collaboration, cybernetic learning models, and responsible AI-driven education.

Keywords: Evolutionary Computation, Complex Network, Application of Artificial Intelligence
Abstract: Structure

This tutorial is organized into five key sections that collectively explore the fusion of evolutionary optimization techniques with both large language models and multimodal LLMs (MLLMs). We begin with an overview of con-ventional evolutionary optimization, laying the groundwork for understanding how EVC tackles complex sys-tems. Next, we introduce LLM-enhanced evolutionary optimization, where the incorporation of MLLM insights further refines traditional methods. The subsequent section examines how evolutionary algorithms can optimise prompt engineering for both LLMs and MLLMs, enhancing their contextual and multimodal outputs. We then demonstrate the integrated synergy of these approaches in revolutionizing code generation. Finally, we conclude with a discussion of future research directions, highlighting emerging opportunities for deeper integration of LLM, MLLM, and EVC in addressing the challenges of systems science and cybernetics.

Outline

This tutorial explores the integration of cutting-edge large language models (LLMs) and multimodal LLMs (MLLMs) with conventional evolutionary optimization techniques to solve complex problems in systems science and cybernetics. By harnessing the domain expertise embedded within LLMs and MLLMs, we enhance evolutionary algorithms (EAs) to drive more informed search strategies, enabling rapid convergence and improved performance in diverse applications such as network routing, resilient architecture design, and automated code generation. Furthermore, the tutorial demonstrates how EAs can optimize prompt engineering for LLMs, fostering a dynamic interplay between natural language understanding and algorithmic refinement. This innovative approach not only paves the way for breakthroughs in human-machine systems but also aligns with the themes of IEEE SMC 2025, emphasizing interdisciplinary methods in cybernetics and complex systems analysis. Attendees will benefit from live demonstrations, comparative case studies, and interactive sessions that collectively underscore the transformative potential of this synergy in tackling real-world challenges.

Keywords: Intelligent Power Grid, Smart Buildings, Smart Cities and Infrastructures, System Modeling and Control
Abstract: Structure

This tutorial will be structured as follows:

• Introduction to Optimization of Complex Systems • Preliminaries on Optimization • Duality and Decomposition Methods • Non-centralized Optimization • Practical Applications

Outline

Optimization and distributed optimization play a key role in modern energy systems, enabling efficient management of resources, coordination of distributed assets, and scalability of control strategies. This tutorial provides an introduction to convex optimization, duality-based decomposition techniques, and distributed optimisation methods with a focus on energy applications. The tutorial will cover theoretical foundations and practical implementations, including parametric optimization, proximal methods, and iterative distributed algorithms. Through motivating examples and practical demonstrations, participants will gain an understanding of the challenges and solutions in energy optimization problems.

Keywords: Intelligent Power Grid, Intelligent Green Production Systems, Trust in Autonomous Systems
Abstract: The global renewable energy transition and electrification will and does require not only technological advancements but also effective communication to foster public understanding and acceptance. That may sound like an empirical statement, but the more important concept is the better that everyone understands the benefits and challenges of transforming the global economy into an electric one, the better the engagement and cooperation the populace will be. Integrating STEM (Science, Technology, Engineering, and Mathematics) disciplines with psychological sciences can enhance the communication and better explain the cost benefit analysis of renewable energy concepts, building consensus across diverse stakeholders and better enable the successful adoption of the global electrification process. This paper explores strategies for leveraging interdisciplinary approaches to address misconceptions, increase engagement, and promote behavioural change. Real-world examples and recommendations are provided to guide this innovative fusion of disciplines. The workshop will focus on Systems Science and Engineering combining skills of psychological science with existing STEM disciplines.

Keywords: Enterprise Information Systems, Decision Support Systems, Distributed Intelligent Systems
Abstract: The advancements of Large language models (LLMs) have provided great opportunities to text-to-SQL tasks to overcome the main challenges to understand complex domain information and complex database structures in business applications. In this paper, we propose a meta-aware learning framework to integrate domain knowledge, database schema, chain-of-thought reasoning processes, and metadata relationships to improve the SQL generation quality. The proposed framework includes four learning strategies: schema-based learning, Chain-of-Thought (CoT) learning, knowledge-enhanced learning, and key information tokenization. This approach provides a comprehensive understanding of database structure and metadata information towards LLM through fine-tuning to improve its performance on SQL generation within business domains. Through two experimental studies, we have demonstrated the superiority of the proposed methods in execution accuracy, multi-task SQL generation capability, and reduction of catastrophic forgetting.

Keywords: Fault Monitoring and Diagnosis, Manufacturing Automation and Systems, System Modeling and Control
Abstract: Accurate fault detection in industrial environments with high-dimensional sensor data presents significant challenges. This paper proposes an explainable AI framework that combines unsupervised deep representation learning with supervised classification for enhanced quality control in smart manufacturing systems. It utilizes a finely tuned deep autoencoder to convert raw data into a compressed latent representation, effectively capturing the underlying structure while removing irrelevant or noisy features. These representations are then used by a downstream classifier to predict faults. Experimental results on a high-dimensional dataset show that the proposed solution outperforms traditional classifiers that process raw features directly. In addition, the framework incorporates an interpretability phase, using a game-theorybased technique to analyze the latent space and identify the most influential features that contribute to faulty predictions.

Keywords: Intelligent Green Production Systems
Abstract: Detecting camouflaged objects in agriculture is a critical computer vision challenge, particularly for green fruits blending with dense foliage. This study introduces DLP-14k, a novel dataset of 14k high-resolution images capturing lime fruits under diverse real-world conditions, including varying lighting, occlusions, and negative samples (e.g., foliage-only images). Unlike general-purpose datasets like COCO or agricultural datasets like PlantDoc, DLP-14k uniquely focuses on camou- flaged fruit detection, enhanced by data augmentation (rotation, flipping, brightness adjustment) to mitigate its small size. We evaluate YOLOv8 and Faster R-CNN for object detection and segmentation, showing YOLOv8’s efficiency for real-time applications but higher false positives in low-contrast settings, while Faster R-CNN achieves superior accuracy in occluded environments at greater computational cost. By introducing DLP-14k and analyzing these trade-offs, this study offers insights for precision agriculture. Future work will expand the dataset and explore hybrid models, advancing applications in complex visual environments.

Keywords: Robotic Systems, Manufacturing Automation and Systems, Quality and Reliability Engineering
Abstract: In the context of Smart Manufacturing, robots are increasingly designed to operate alongside humans, with collaborative robots playing a central role. Ensuring safety in such Human–Robot Collaboration (HRC) scenarios require advanced path planning systems capable of detecting and responding to obstacles, including humans, in real time, thereby enabling safe and efficient cooperation. This paper presents a comparative study of optimization-based path planning techniques applied in collaborative robotics for obstacle avoidance. A structured comparison is conducted among different deterministic optimization strategies, such as Newton’s Method, Conjugate Gradient, and Gradient Descent, each employing various methods for obstacle pose identification, estimation, and danger factor modeling. Through an extensive review of these methods and their outcomes, the study evaluates their performance in terms of path safety, computational efficiency, and adaptability to dynamic environments. The analysis highlights the strengths and limitations of each optimization-based model and provides guidance for selecting suitable path planning approaches for different robotic applications.

Keywords: System Modeling and Control, Adaptive Systems, Fault Monitoring and Diagnosis
Abstract: This paper introduces a new Retrospective Cost Input Estimation (RCIE) method for disturbance estimation and modeling in Intelligent Manufacturing Systems (IMS). Key contributions include an estimation method that works with cascaded subsystems and nonlinearities, regardless of plant zero types, and a framework that combines RCIE with the Equivalent-Input-Disturbance (EID) estimator for improved performance, especially in systems with poor observability. The effectiveness of RCIE is validated through simulations and physical experiments on a motor system, showing it outperforms the Augmented Kalman Filter (AKF) in accuracy and convergence, particularly with non-sinusoidal inputs and complex dynamics. It is the first to the authors knowledge to highlight the connection between an Equivalent Input disturbance and retrospective cost optimisation. This RCIE-based approach offers a robust solution for state and input estimation in IMS, with potential for better control, fault diagnosis, and manufacturing resilience.

Keywords: Digital Twin, Robotic Systems, Control of Uncertain Systems
Abstract: The objective of this work is to create a digital twin of a lower-limb exoskeleton utilizing data from an inertial measurement unit (IMU) to monitor and assess mobility in real time. The virtual model, developed in MATLAB Simscape, utilizes In- ertial measurement unit (IMU) inputs from the lower limbs to replicate the user’s joint movements. This enables the examination of kinematics, validation of performance, and assessment of control without the necessity of repeatedly conducting the same physical tests. This method demonstrates the feasibility of conducting gait analysis and enhancing exoskeletons in a manner that is both practical and economical.

Keywords: Application of Artificial Intelligence, Cybernetics for Informatics, Biometric Systems and Bioinformatics
Abstract: Voice-based diagnostic systems offer a scalable solution for remote health assessment. However, recent advances in generative voice models may enable malicious manipulation of voice samples to simulate or conceal disease-related speech characteristics, which poses new risks to diagnostic systems. This paper investigates the vulnerability of diagnostic and detection models to such types of ``deepfake'' attacks. We show that it is possible to train a generative model to convert between healthy voices and pathological ones, which in turn, can successfully deceive existing diagnostic systems. Here, focus is placed on COVID-19 infection and respiratory abnormalities, but the method can be applied across different pathological conditions affecting vocal attributes. We also benchmark four state-of-the-art synthesized voice detection models on both real and generated pathological speech from three datasets. Our results show that current synthetic voice detectors, typically trained on healthy speech data, perform poorly on generated pathological samples. While fine-tuning with real pathological voices improves detection, a substantial performance gap remains. This work provides initial insights on an emerging threat to remote voice diagnostic systems that needs further work.

Keywords: Artificial Social Intelligence, Multimedia Computation, Deep Learning
Abstract: With the emergence of multi-modal generative models, synthesized videos are becoming increasingly realistic, making the detection of deepfakes extremely challenging. While several video deepfake detection models have shown promising performance, their focus has been primarily on the visual modality. To overcome this limitation, we propose a dual- stream framework that fuses visual and auditory information via cross-attention computed between embeddings extracted from pre-trained video and audio encoders. Additionally, we design a weakly-supervised forgery localization head that infers frame-level forgery scores from coarse segment-level labels, minimizing the need for fine-grained annotations and allowing for forgery location characterization. In this paper, we describe our preliminary results showing the proposed model outperforming state-of-the-art detectors on both frame- level localization and sequence-level deepfake detection tasks. Ongoing work focuses on investigating the complementarity between the visual and auditory modalities to improve model robustness and explainability.

Keywords: Computational Intelligence, Hybrid Models of Neural Networks, Fuzzy Systems, and Evolutionary Computing, Deep Learning
Abstract: Retrieval-Augmented Generation (RAG) was first introduced to enhance the capabilities of Large Language Models (LLMs) beyond their own encoded-prior knowledge. This is possible by making available to the LLMs an external source of knowledge, which helps to reduce factual hallucinations and enables the access to new information, typically not available during the training phase of LLMs. Despite its benefits, there is an increasing concern with the impact of inconsistent retrieved information towards LLMs’ responses. Hence, the Retrieval- Augmented Generation Benchmark (RGB) was introduced as a new testbed for RAG evaluation, meant to assess the robustness of LLMs towards inconsistency in the retrieved information received from the external knowledge. In this work, we use the RGB corpus to evaluate LLMs in four scenarios: (1) noise robustness; (2) information integration; (3) negative rejection; and (4) counterfactual robustness. We perform a comparative analysis between the RAG baseline defined by the RGB and variations of GraphRAG, a Knowledge Graph (KG) based RAG system developed to retrieve relevant information from large text. We tested GraphRAG under three customization to improve its robustness. Our approach demonstrates improvements compared to the RGB baseline system, providing insights on how to design more reliable RAG systems, tailored for real-world scenarios.

Keywords: Representation Learning, Deep Learning, AI and Applications
Abstract: The growing sophistication of speech generated by Artificial Intelligence (AI) has introduced new challenges in audio deepfake detection. Text-to-speech (TTS) and voice con version (VC) technologies can now produce convincing synthetic speech with high quality and intelligibility. This poses a serious threat to voice biometric security systems, such as automatic speaker recognition. It also increases the risks associated to the spread of spoken disinformation, where synthetic voices can be used to disseminate malicious content. In this study, we conduct an analysis of real and synthetic speech at phonetic and word levels. For that, a parallel dataset comprising real and synthetic speech signals were developed based on a subset of the LibriSpeech ASR corpus. Synthetic speech samples were generated using two TTS and one VC systems: Coqui TTS, VITS TTS, and StarGANv2 VC. We adopted HuBERT, a self-supervised speech model, to extract speech embeddings. The motivation for using this model stems from its ability to recognize sound units corresponding to the so-called pseudo phonemes. Our analysis is based on the KL divergence (KLD) between the distributions of synthetic and real phonemes, which allowed us to rank synthetic phonemes based on their alignment with their real counterpart. We also trained several classifiers per phoneme to distinguish between real and synthetic samples. We then compute the correlations between KLD and accuracies per phoneme. Besides showing a list of phonemes that are more discriminative, our findings suggest that vowels correlate better with the classifiers’ performance, suggesting that the KLD can be an indicator of the most distinguishable phonemes for deepfake detection.

Keywords: Quantum Cybernetics, Quantum Machine Learning
Abstract: Structure

The tutorial will be organized in two main parts. The first part will be devoted to introducing the quantum computational intelligence research area including basic concepts of quantum computing. The second part, instead, will consider the use of quantum computing to enhance computational intelligence techniques. For this reason, this part of the tutorial will discuss the implementation of a genetic algorithm, a fuzzy system and neural networks on a quantum computer. Outline

Computational Intelligence and Quantum Computing are two highly topical fields of research that can benefit from each other’s discoveries by opening a completely new scenario in computation, that of quantum computational intelligence. Indeed, on the one hand, computational intelligence algorithms can be made computationally more efficient due to the massive parallelism induced by quantum phenomena; on the other hand, the complex development of quantum computing technologies and methodologies can be properly supported by the use of classical computational intelligence approaches. In the area of quantum computational intelligence, this tutorial will focus on discussing how computational intelligence techniques such as genetic algorithms, neural networks and fuzzy systems can be implemented with the support of quantum computers. From a cybernetics perspective, Quantum Computational Intelligence represents an evolution in how intelligent systems process information and make decisions. Specifically, this tutorial will allow auditors with no particular knowledge of quantum computing to acquire information on the meaning of the term quantum computational intelligence, the main approaches developed in this field of research such as quantum evolutionary algorithms, quantum fuzzy systems and quantum neural networks, the tools used to implement quantum algorithms and, finally, the challenges that need to be addressed in order to consolidate this field of research.

Keywords: Passive BMIs, Active BMIs, BMI Emerging Applications
Abstract: Further details are provided at https://www.ieeesmc2025.org/br41n-io-designers-hackathon.html and https://www.br41n.io/IEEE-SMC-2025